Run a fleet of coding agents

Spawn, watch, approve, and orchestrate AI coding agents — from one cockpit.

Get started ★ Star on GitHub

brew install --cask srjn45/tap/warden

Works with Claude Code, Codex, Antigravity, Cursor, Aider, OpenCode, Crush, and Goose.

warden start and warden ls in a terminal

🧭 One cockpit

TUI + web. See every agent's status at a glance.

🔗 Pipelines

Chain dependent agents into a DAG — analyze → implement → review.

💸 Measured savings

A real ledger of tokens kept out of context, plus per-model cost governance.

📦 One binary

Daemon, CLI, TUI, web, MCP. No database.

The TUI cockpit

Framed panes, live agent detail, scrollable sessions, approvals inbox — all in the terminal.

Web mission control

A URL-routed dashboard with a Cockpit home, live SSE fleet, grouping, and a one-click attention queue — in the browser.

Fleet metrics & savings

CPU, memory, context fill and fleet size per agent and fleet-wide — plus the tokens-saved trend and per-model cost, from the savings ledger.

Everything warden does

Spawn & classify Approvals & auto-approve Supervised mode Pipelines (DAG)Worktrees & task types Self-rotation & digests Savings & cost governance Context guard & compaction Snapshots & rollback Scheduling Remote access Insights

Pipelines

Chain dependent agents into a DAG — outputs flow downstream, fan-out and fan-in resolve automatically, and the web UI renders the live graph.

A multi-stage agent pipeline rendered as a DAG in the warden web UI: parallel roots fan out through design and implementation jobs and converge on integrate, test, review and release jobs.

Quickstart

brew install --cask srjn45/tap/warden   # or deb / rpm / AUR / source — see Install
warden doctor          # preflight checks
warden start "fix the flaky nightly test"